XGBoost based residual life prediction in the presence of human error in maintenance

Accurate maintenance decision making is essential for organizations like military and aviation. Immensely demanding situations like limited time availability for maintenance in strenuous conditions escalate the possibility of human errors in maintaining such equipment. Human errors in maintenance negatively impact the life of the systems. Human Reliability Analysis methodologies have evolved to systematically quantify the human error in terms of Human Error Probability. However, the exact effect of human error on every component’s life is unknown yet. In the presence of the diverse operating profiles for equipment, estimating such effects becomes a complex and mathematically challenging problem to be handled by conventional statistical techniques. This paper presents a machine learning approach to estimate the residual life of a component by incorporating the effect of human error in maintenance. Based on the nature of the maintenance data, a gradient boosting ensemble model (XGBoost) is developed, which predicts the residual life of the component while considering error induced by maintenance personnel during its maintenance. The model recommends the maintenance decision considering the predicted residual life and the user-defined future mission profile. Additionally, provision is made to capture the stochastic future operating profile. The developed model effectively handles the uncertainties and variabilities in expected future mission profiles and the correlation of multiple influencing parameters without increasing mathematical complexity. The developed model is illustrated in the decision making of replacement of a component in a mission-critical military system in pre-mission maintenance break. From the perspective of managerial implications, some of the key findings from numerical experiments on the developed model are presented.